What each AI browser sends, and to whom
Short answer
Every hosted assistant sends the page you are viewing to its vendor; what differs is retention and training use. Brave says it keeps nothing and trains on nothing, Microsoft says it may train on your data, and Google says some of the page content it uses is not visible to you.
The one thing they have in common
A browser assistant that answers on a server has to receive the page in order to answer. That is not a flaw in any of these products, it is how they work, and no setting changes it.
So the question is never whether the page is sent. It is what happens to it afterwards, and the vendors differ enormously on that.
Where they actually differ
Brave Leo is the strongest hosted position we found: Brave's notice says it does not store or retain prompts, responses or context, does not use conversations for training, and does not collect identifiers linkable to you.
Copilot in Edge is the most candid in the other direction. Microsoft's privacy statement says it may use your data to develop, train and fine-tune its AI models, including large language models.
Gemini in Chrome collects the page content and URL of your current tab, and Google's own notice says some of the page content Gemini uses might not be visible to you. With Keep Activity on, that information is retained in Gemini Apps Activity and used to improve services with the help of human reviewers.
Perplexity's notice lists improving and creating its AI models among its uses of collected data, and carves out the Email Assistant specifically as not used for training — which tells you what the default looks like.
Dia's policy says content may be used to improve Dia unless you opt out, and gives no retention period for it.
Firefox is the odd one out: translation and PDF alt-text run on-device with small models, and the chatbot sidebar is a shell around a provider you choose, so Mozilla says it does not see the conversation and the provider does.
How to read these documents yourself
Look for the carve-outs. When a policy names one feature as exempt from training, it is describing the rule for everything else.
Look for a retention period, in days. “As long as necessary to provide the service” is not a period.
Separate “we do not sell your data” from “we do not retain it” from “it never leaves your device”. Only the last can be checked from outside, and it is checked with a network monitor rather than with a lawyer.
The dates matter
Every statement above comes from the vendor's own documentation, read on 4 November 2026, with the source linked on each browser's page. These documents change — that is the point of writing the date down, and the reason to check the current version before relying on any of it.
Other ways people ask this
- does my browser send data to ai company
Related
- Up a levelGuides: local AI, browser privacy and the hardware to run it
- Closely relatedAI assistants and your banking pages
- Closely relatedWhat Brave Leo sends to Brave, and how to check
- Closely relatedChatGPT Atlas has been retired: what happened to your data
- Elsewhere on the siteAI browser alternatives, compared honestly
- Elsewhere on the siteAI browsers compared: where each one sends your pages